Large ARFs increase renaming cost

Article: ARM Goes 64-bit
By: Paul A. Clayton (paaronclayton.delete@this.gmail.com), November 17, 2012 9:23 pm
Room: Moderated Discussions
name99 (name99.delete@this.redheron.com) on November 17, 2012 6:37 pm wrote:
[snip]
> Let's answer slightly differently.
> If you have a modern type of CPU (OoO, superscalar, prediction, all that good stuff) then there is a distinction
> between the number of PHYSICAL registers and the number of ARCHITECTED (ie expressible in assembly language)
> registers. The number of physical registers is going to be determined by how aggressively OoO you want your CPU
> to be, and it is the number of physical registers that determines power, area, cycle times and so on.
>
> Given this, the number of architected registers is essentially free. Going from 16 to 32 costs you
> an extra bit in each register specification (so 3 bits in most instructions) and that's it. If there
> is any advantage to increasing the number of architected registers, you might as well do so.

In an OoO implementation, one will also have a renaming table in the front end (and often one for commit as well, but that table has lesser requirements). Increasing the size of this table will increase power use and may complicate pipelining.

In addition, with a large number of architectural registers (with an implicit flat organization) many registers may be relatively idle but consume the same static power and increase the area of the register file (and so the latency and power of all register accesses).

Scalability of an ISA is also important. A large (flat) architectural register file will be less friendly to smaller implementations. (Such will also be less friendly to multithreading.) A 64 GPR ISA would require 33% more physical registers for GPRs than a 32 GPR ISA with 64 rename registers (which could satisfy an 80 instruction window in many cases). (There are also techniques to reduce the number of physical rename register required which do not apply as much or at all to architectural registers, e.g., virtual physical registers.)

> A second way to think about this is in terms of this as a "power" feature, like vector instructions, that will
> be used by people who know what they are doing, and not otherwise. For example: people who know what they are
> doing, on entry to a function, IMMEDIATELY load into a local (ie register) variable all globals that will be
> accessed, along with everything that will be used through a pointer. They do all their calculation in the local
> variables, then store everything on exit from the function. This style of coding requires a lot more registers
> to work with, but is also faster. (It's faster because it gets load latency off the critical path, and because
> it doesn't waste cycles doing things the compiler thinks might be necessary --- writing back globals, writing
> back instance variables --- but which you know are not, every time they are changed.)

In general high level language code will not express register allocation. With the increased interest in interprocedural and whole-program optimization, compilers should be fairly well equipped to exploit more registers.

It should also be noted that loading all registers at the start of a function may well not be an optimization. Even with an OoO implementation, distributing memory accesses through code tends to be more execution friendly.

> Why isn't the code you profiled showing this sort of thing? The cruel answer would be that there are just
> not that many people in the world who know what they are doing. A second alternative, which might be partially
> true, is that the bulk of programmers, and the bulk of code written, grew up with IA-32, not even x64,
> where this type of programming is not much of a win because of the paucity of registers. Ideally this will
> change, but we all know that it takes a generation or more for certain habits to die.

Or the bulk of code is written in a C-like language (with aliasing issues) and written for the ease of programming (at minimum writing, more ideally for ease of maintenance).

> Which gets back to my point. Plough-the-fields code will tend not to use half those registers, just
> like it doesn't use multi-threading or NEON. But performance critical code WILL use these features.

Compilers are becoming more sophisticated in autovectorization and register allocation may be an easier and more mature optimization. Autoparallelization is also becoming more common for certain types of programs.
< Previous Post in ThreadNext Post in Thread >
TopicPosted ByDate
New Article: ARM Goes 64-bitDavid Kanter2012/08/14 12:04 AM
  New Article: ARM Goes 64-bitnone2012/08/14 12:44 AM
    New Article: ARM Goes 64-bitDavid Kanter2012/08/14 01:04 AM
    MIPS MT-ASEPaul A. Clayton2012/08/14 09:01 AM
      MONITOR/MWAITEduardoS2012/08/14 10:08 AM
        MWAIT not specifically MTPaul A. Clayton2012/08/14 10:36 AM
          MWAIT not specifically MTEduardoS2012/08/15 03:16 PM
        MONITOR/MWAITanonymou52012/08/14 11:07 AM
          MONITOR/MWAITEduardoS2012/08/15 03:20 PM
      MIPS MT-ASErwessel2012/08/14 10:14 AM
  New Article: ARM Goes 64-bitSHK2012/08/14 02:01 AM
  New Article: ARM Goes 64-bitanon2012/08/14 02:37 AM
    New Article: ARM Goes 64-bitRichard Cownie2012/08/14 03:57 AM
      New Article: ARM Goes 64-bitanon2012/08/14 04:29 AM
      New Article: ARM Goes 64-bitnone2012/08/14 04:44 AM
        New Article: ARM Goes 64-bitanon2012/08/14 05:28 AM
          New Article: ARM Goes 64-bitanon2012/08/14 05:32 AM
            New Article: ARM Goes 64-bitEduardoS2012/08/14 06:06 AM
          New Article: ARM Goes 64-bitnone2012/08/14 05:40 AM
            AArch64 select better than cmovPaul A. Clayton2012/08/14 06:08 AM
            New Article: ARM Goes 64-bitanon2012/08/14 06:12 AM
              New Article: ARM Goes 64-bitnone2012/08/14 06:25 AM
                Predicated ld/store are usefulPaul A. Clayton2012/08/14 06:48 AM
                  Predicated ld/store are usefulnone2012/08/14 06:56 AM
                    Predicated ld/store are usefulanon2012/08/14 07:07 AM
                    Predicated stores might not be that badPaul A. Clayton2012/08/14 07:27 AM
                      Predicated stores might not be that badDavid Kanter2012/08/15 01:14 AM
                        Predicated stores might not be that badMichael S2012/08/15 11:41 AM
                        Predicated stores might not be that badR Byron2012/08/17 04:09 AM
                New Article: ARM Goes 64-bitanon2012/08/14 06:54 AM
                  New Article: ARM Goes 64-bitnone2012/08/14 07:04 AM
                    New Article: ARM Goes 64-bitanon2012/08/14 07:43 AM
          New Article: ARM Goes 64-bitEduardoS2012/08/14 06:07 AM
            New Article: ARM Goes 64-bitanon2012/08/14 06:20 AM
              New Article: ARM Goes 64-bitnone2012/08/14 06:29 AM
                New Article: ARM Goes 64-bitanon2012/08/14 07:00 AM
            New Article: ARM Goes 64-bitMichael S2012/08/14 03:43 PM
        New Article: ARM Goes 64-bitRichard Cownie2012/08/14 06:53 AM
          OT: Conrad's "Youth"Richard Cownie2012/08/14 07:20 AM
      New Article: ARM Goes 64-bitEduardoS2012/08/14 06:04 AM
        New Article: ARM Goes 64-bitmpx2012/08/14 08:59 AM
          New Article: ARM Goes 64-bitAntti-Ville Tuunainen2012/08/14 09:16 AM
        New Article: ARM Goes 64-bitanonymou52012/08/14 11:03 AM
          New Article: ARM Goes 64-bitname992012/11/17 03:31 PM
            Microarchitecting a counter registerPaul A. Clayton2012/11/17 07:37 PM
    New Article: ARM Goes 64-bitbakaneko2012/08/14 04:21 AM
      New Article: ARM Goes 64-bitname992012/11/17 03:40 PM
        New Article: ARM Goes 64-bitEduardoS2012/11/17 04:52 PM
        New Article: ARM Goes 64-bitDoug S2012/11/17 05:48 PM
        New Article: ARM Goes 64-bitbakaneko2012/11/18 05:40 PM
          New Article: ARM Goes 64-bitWilco2012/11/19 07:59 AM
            New Article: ARM Goes 64-bitEduardoS2012/11/19 08:23 AM
              New Article: ARM Goes 64-bitWilco2012/11/19 09:31 AM
                Downloading µarch-specific binaries?Paul A. Clayton2012/11/19 11:21 AM
                New Article: ARM Goes 64-bitEduardoS2012/11/19 11:41 AM
                  New Article: ARM Goes 64-bitWilco2012/11/21 07:44 AM
                    JIT vs. static compilation (Was: New Article: ARM Goes 64-bit)VMguy2012/11/22 03:21 AM
                      JIT vs. static compilation (Was: New Article: ARM Goes 64-bit)David Kanter2012/11/22 12:12 PM
                        JIT vs. static compilation (Was: New Article: ARM Goes 64-bit)Gabriele Svelto2012/11/23 03:50 AM
                    New Article: ARM Goes 64-bitEduardoS2012/11/23 10:09 AM
                      New Article: ARM Goes 64-bitEBFE2012/11/26 01:24 AM
                        New Article: ARM Goes 64-bitGabriele Svelto2012/11/26 03:33 AM
                          New Article: ARM Goes 64-bitEBFE2012/11/27 11:17 PM
                            New Article: ARM Goes 64-bitGabriele Svelto2012/11/28 02:32 AM
                        New Article: ARM Goes 64-bitEduardoS2012/11/26 12:16 PM
                          New Article: ARM Goes 64-bitEBFE2012/11/28 12:33 AM
                            New Article: ARM Goes 64-bitEduardoS2012/11/28 05:53 AM
                              New Article: ARM Goes 64-bitMichael S2012/11/28 06:15 AM
                                New Article: ARM Goes 64-bitEduardoS2012/11/28 07:33 AM
                                  New Article: ARM Goes 64-bitMichael S2012/11/28 09:16 AM
                                    New Article: ARM Goes 64-bitEduardoS2012/11/28 09:53 AM
                                    New Article: ARM Goes 64-bitEugene Nalimov2012/11/28 05:58 PM
                                      Amazing!EduardoS2012/11/28 07:25 PM
                                        Amazing! (non-italic response)EduardoS2012/11/28 07:25 PM
                                        Amazing!EBFE2012/11/28 08:20 PM
                                          Undefined behaviour doubles downEduardoS2012/11/28 09:10 PM
                              New Article: ARM Goes 64-bitEBFE2012/11/28 07:54 PM
                                New Article: ARM Goes 64-bitEduardoS2012/11/28 09:21 PM
                Have you heard of Transmeta?David Kanter2012/11/19 03:47 PM
            New Article: ARM Goes 64-bitbakaneko2012/11/19 09:08 AM
            New Article: ARM Goes 64-bitDavid Kanter2012/11/19 03:40 PM
              Semantic Dictionary EncodingRay2012/11/19 10:37 PM
              New Article: ARM Goes 64-bitRohit2012/11/20 04:48 PM
                New Article: ARM Goes 64-bitDavid Kanter2012/11/20 11:07 PM
                  New Article: ARM Goes 64-bitWilco2012/11/21 06:41 AM
                    New Article: ARM Goes 64-bitDavid Kanter2012/11/21 10:12 AM
                    A JIT exampleMark Roulo2012/11/21 10:30 AM
                      A JIT exampleWilco2012/11/21 07:04 PM
                        A JIT examplerwessel2012/11/21 09:05 PM
                        A JIT exampleGabriele Svelto2012/11/23 03:53 AM
                        A JIT exampleEduardoS2012/11/23 10:13 AM
                          A JIT exampleWilco2012/11/23 01:41 PM
                            A JIT exampleEduardoS2012/11/23 02:06 PM
                            A JIT exampleGabriele Svelto2012/11/23 04:09 PM
                              A JIT exampleSymmetry2012/11/26 05:58 AM
            New Article: ARM Goes 64-bitRay2012/11/19 10:27 PM
    New Article: ARM Goes 64-bitDavid Kanter2012/08/14 09:11 AM
  v7-M is Thumb-onlyPaul A. Clayton2012/08/14 06:58 AM
  Minor suggested correctionPaul A. Clayton2012/08/14 08:33 AM
    Minor suggested correctionanon2012/08/14 08:57 AM
  New Article: ARM Goes 64-bitExophase2012/08/14 08:33 AM
    New Article: ARM Goes 64-bitDavid Kanter2012/08/14 09:16 AM
      New Article: ARM Goes 64-bitjigal2012/08/15 01:49 PM
  Correction re ARM and BBC MicroPaul2012/08/14 08:59 PM
    Correction re ARM and BBC MicroPer Hesselgren2012/08/15 03:27 AM
  Memory BW so lowPer Hesselgren2012/08/15 03:14 AM
    Memory BW so lownone2012/08/15 11:16 AM
  New Article: ARM Goes 64-bitdado2012/08/15 10:25 AM
  Number of GPRsKenneth Jonsson2012/08/16 02:35 PM
    Number of GPRsExophase2012/08/16 02:52 PM
      Number of GPRsKenneth Jonsson2012/08/17 02:41 AM
        Ooops, missing link...Kenneth Jonsson2012/08/17 02:44 AM
        64-bit pointers eat some performancePaul A. Clayton2012/08/17 06:19 AM
          64-bit pointers eat some performancebakaneko2012/08/17 08:37 AM
            Brute force seems to workPaul A. Clayton2012/08/17 10:08 AM
              Brute force seems to workbakaneko2012/08/17 11:15 AM
          64-bit pointers eat some performanceRichard Cownie2012/08/17 08:46 AM
            Pointer compression is atypicalPaul A. Clayton2012/08/17 10:43 AM
              Pointer compression is atypicalRichard Cownie2012/08/17 12:57 PM
                Pointer compression is atypicalHoward Chu2012/08/22 10:17 PM
                  Pointer compression is atypicalRichard Cownie2012/08/23 04:48 AM
                    Pointer compression is atypicalHoward Chu2012/08/23 06:51 AM
              Pointer compression is atypicalWilco2012/08/17 02:41 PM
                Pointer compression is atypicalRichard Cownie2012/08/17 04:13 PM
                  Pointer compression is atypicalRicardo B2012/08/19 10:44 AM
                  Pointer compression is atypicalHoward Chu2012/08/22 10:08 PM
                    Unified libraries?Paul A. Clayton2012/08/23 07:49 AM
                    Pointer compression is atypicalRichard Cownie2012/08/23 08:44 AM
                      Pointer compression is atypicalHoward Chu2012/08/23 05:17 PM
                        Pointer compression is atypicalanon2012/08/23 08:15 PM
                          Pointer compression is atypicalHoward Chu2012/08/23 09:33 PM
            64-bit pointers eat some performanceFoo_2012/08/18 12:09 PM
              64-bit pointers eat some performanceRichard Cownie2012/08/18 05:25 PM
                64-bit pointers eat some performanceRichard Cownie2012/08/18 05:32 PM
            Page-related benefit of small pointersPaul A. Clayton2012/08/23 08:36 AM
        Number of GPRsWilco2012/08/17 06:31 AM
          Number of GPRsKenneth Jonsson2012/08/17 11:54 AM
            Number of GPRsExophase2012/08/17 12:44 PM
              Number of GPRsKenneth Jonsson2012/08/17 01:22 PM
                Number of GPRsWilco2012/08/17 02:53 PM
        What about dynamic utilization?Exophase2012/08/17 09:30 AM
          Compiler vs. assembly aliasing knowledge?Paul A. Clayton2012/08/17 10:20 AM
            Compiler vs. assembly aliasing knowledge?Exophase2012/08/17 11:09 AM
            Compiler vs. assembly aliasing knowledge?anon2012/08/18 02:23 AM
              Compiler vs. assembly aliasing knowledge?Ricardo B2012/08/19 11:02 AM
                Compiler vs. assembly aliasing knowledge?anon2012/08/19 06:07 PM
                  Compiler vs. assembly aliasing knowledge?Ricardo B2012/08/19 07:26 PM
                    Compiler vs. assembly aliasing knowledge?anon2012/08/19 10:03 PM
                      Compiler vs. assembly aliasing knowledge?anon2012/08/20 01:59 AM
        Number of GPRsDavid Kanter2012/08/17 12:46 PM
          RAT issues as part of reason 1Paul A. Clayton2012/08/17 02:18 PM
        Number of GPRsname992012/11/17 06:37 PM
          Large ARFs increase renaming costPaul A. Clayton2012/11/17 09:23 PM
    Number of GPRsDavid Kanter2012/08/16 03:31 PM
    Number of GPRsRichard Cownie2012/08/16 05:17 PM
    32 GPRs ~2-3%Paul A. Clayton2012/08/16 06:27 PM
      Oops, Message-ID: aaed6e38-c7bd-467e-ba41-f40cf1020e5e@googlegroups.com (NT)Paul A. Clayton2012/08/16 06:29 PM
      32 GPRs ~2-3%Exophase2012/08/16 10:06 PM
        R31 as SP/zero is kind of neat (NT)Paul A. Clayton2012/08/17 06:23 AM
        32 GPRs ~2-3%rwessel2012/08/17 08:24 AM
          32 GPRs ~2-3%Exophase2012/08/17 09:16 AM
            32 GPRs ~2-3%Max2012/08/17 04:19 PM
      32 GPRs ~2-3%name992012/11/17 07:43 PM
    Number of GPRsmpx2012/08/17 01:11 AM
      Latency and powerPaul A. Clayton2012/08/17 06:54 AM
    Number of GPRsbakaneko2012/08/17 03:09 AM
  New Article: ARM Goes 64-bitSteve2012/08/17 02:12 PM
    New Article: ARM Goes 64-bitDavid Kanter2012/08/19 12:42 PM
      New Article: ARM Goes 64-bitDoug S2012/08/19 02:02 PM
      New Article: ARM Goes 64-bitAnon2012/08/19 07:16 PM
      New Article: ARM Goes 64-bitSteve2012/08/30 07:51 AM
  Scalar vs Vector registersRobert David Graham2012/08/19 05:19 PM
    Scalar vs Vector registersDavid Kanter2012/08/19 05:29 PM
  New Article: ARM Goes 64-bitBaserock ARM servers2012/08/21 04:13 PM
    Baserock ARM serversSysanon2012/08/21 04:14 PM
    A-15 virtualization and LPAE?Paul A. Clayton2012/08/21 06:13 PM
      A-15 virtualization and LPAE?Anon2012/08/21 07:13 PM
        Half-depth advantages?Paul A. Clayton2012/08/21 08:42 PM
          Half-depth advantages?Anon2012/08/22 03:33 PM
            Thanks for the information (NT)Paul A. Clayton2012/08/22 04:04 PM
      A-15 virtualization and LPAE?C. Ladisch2012/08/23 11:12 AM
        A-15 virtualization and LPAE?Paul2012/08/23 03:17 PM
        Excessive pessimismPaul A. Clayton2012/08/23 04:08 PM
          Excessive pessimismDavid Kanter2012/08/23 05:05 PM
    New Article: ARM Goes 64-bitMichael S2012/08/22 07:12 AM
      BTW, Baserock==product, Codethink==company (NT)Paul A. Clayton2012/08/22 08:56 AM
  New Article: ARM Goes 64-bitReinoud Zandijk2012/08/21 11:27 PM
Reply to this Topic
Name:
Email:
Topic:
Body: No Text
How do you spell green?